U.S. patent application number 11/104831 was filed with the patent office on 2005-12-08 for method and system for registering pre-procedural images with intra-procedural images using a pre-computed knowledge base.
Invention is credited to Khamene, Ali, Navab, Nassir, Sauer, Frank, Xu, Chenyang.
Application Number | 20050272991 11/104831 |
Document ID | / |
Family ID | 35160459 |
Filed Date | 2005-12-08 |
United States Patent
Application |
20050272991 |
Kind Code |
A1 |
Xu, Chenyang ; et
al. |
December 8, 2005 |
Method and system for registering pre-procedural images with
intra-procedural images using a pre-computed knowledge base
Abstract
A system and method for registering pre-operative images of an
object with an intra-operative image of the object is disclosed.
Prior to an operative procedure, Digitally Reconstructed
Radiographs (DRRs) are generated for the pre-operative images of
each individual patient. Signatures are extracted from the DRRs.
The signatures are stored in a knowledge base. During the operative
procedure, a signature is extracted from the intra-operative image.
The intra-operative signature is compared to the stored
pre-operative signatures. A pre-operative image having a best
signature match to the intra-operative signature is retrieved. The
retrieved pre-operative image is registered with the
intra-operative image.
Inventors: |
Xu, Chenyang; (Allentown,
NJ) ; Navab, Nassir; (Plainsboro, NJ) ; Sauer,
Frank; (Princeton, NJ) ; Khamene, Ali;
(Princeton, NJ) |
Correspondence
Address: |
SIEMENS CORPORATION
INTELLECTUAL PROPERTY DEPARTMENT
170 WOOD AVENUE SOUTH
ISELIN
NJ
08830
US
|
Family ID: |
35160459 |
Appl. No.: |
11/104831 |
Filed: |
April 13, 2005 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
60564508 |
Apr 22, 2004 |
|
|
|
Current U.S.
Class: |
600/407 ;
600/416; 600/423 |
Current CPC
Class: |
A61B 6/032 20130101;
A61B 5/055 20130101 |
Class at
Publication: |
600/407 ;
600/416; 600/423 |
International
Class: |
A61B 005/05 |
Claims
We claim:
1. A method for registering pre-operative images of an object with
an intra-operative image of the object comprising the steps of:
prior to an operative procedure, generating Digitally Reconstructed
Radiographs (DRRs) for the pre-operative images at various poses
for each patient; extracting signatures from the DRRs; storing the
signatures in a knowledge base; during the operative procedure,
extracting a signature from the intra-operative image; comparing
the intra-operative signature to the stored pre-operative
signatures; retrieving a pre-operative image having a best
signature match to the intra-operative signature; and registering
the retrieved pre-operative image with the intra-operative
image.
2. The method of claim 1 wherein the pre-operative images are three
dimensional (3D) images.
3. The method of claim 2 wherein the pre-operative images are
computed tomography images.
4. The method of claim 2 wherein the pre-operative images are
magnetic resonance images.
5. The method of claim 1 wherein the intra-operative image is a two
dimensional (2D) image.
6. The method of claim 5 wherein the intra-operative image is an
X-ray image.
7. The method of claim 5 wherein the intra-operative image is a
fluoroscopy image.
8. The method of claim 1 wherein at least one signature is a
histogram.
9. The method of claim 1 wherein at least one signature is an
invariant derived from multiscale Gaussian filters.
10. The method of claim 1 wherein at least one signature is an
invariant derived from multiscale Gabor filters.
11. The method of claim 1 wherein a distance measurement is used to
compare the intra-operative signature to the pre-operative
signatures.
12. The method of claim 1 wherein the object is a patient.
13. A system for registering pre-operative images of an object with
an intra-operative image of the object comprising: a two
dimensional (2D) imaging system; a processor for performing the
following steps: i). receiving pre-operative images generated by a
three dimensional (3D) imaging system; ii). generating Digitally
Reconstructed Radiographs (DRRs) of the pre-operative images; iii).
receiving inter-operative images generated by the 2D system and
extracting signatures from the images; iv). extracting signatures
from the DRRs; v). comparing the intra-operative signature to the
pre-operative signatures; vi). identifying a pre-operative image
having a best signature match to the intra-operative signature; and
vii). registering the identified pre-operative image with the
intra-operative image; a database for storing the pre-operative,
inter-operative and registered images; and a display for displaying
the images.
14. The system of claim 13 wherein the pre-operative images are
computed tomography images.
15. The system of claim 13 wherein the pre-operative images are
magnetic resonance images.
16. The system of claim 13 wherein the intra-operative image is an
X-ray image.
17. The system of claim 13 wherein the intra-operative image is a
fluoroscopy image.
18. The system of claim 13 wherein at least one signature is a
histogram.
19. The system of claim 13 wherein at least one signature is an
invariant derived from multiscale Gaussian filters.
20. The system of claim 13 wherein at least one signature is an
invariant derived from multiscale Gabor filters.
21. The system of claim 13 wherein a distance measurement is used
to compare the intra-operative signature to the pre-operative
signatures.
22. The system of claim 13 wherein the object is a patient.
Description
CROSS REFERENCE TO RELATED APPLICATIONS
[0001] This application claims the benefit of U.S. Provisional
Patent Application Ser. No. 60/564,508, filed Apr. 22, 2004, which
is incorporated by reference in its entirety.
FIELD OF THE INVENTION
[0002] The present invention is directed to a method and system for
registering pre-procedural images with intra-procedural images, and
more particularly, to a system and method for registering
pre-procedural images with intra-procedural images using a
pre-computed knowledge base of the pre-procedural image data.
BACKGROUND OF THE INVENTION
[0003] Medical professionals have recently been exploiting
pre-operative or pre-procedural images and intra-operative or
intra-procedural images to provide a more useful and inexpensive
registered image of an organ, which is the subject of a minimally
invasive therapeutic intervention. For example, a tumor can be
imaged both pre-operatively using a CT system and intra-operatively
using an X-ray system. Digital Reconstructed Radiographs (DRRs) are
reconstructed from the CT images to model the X-ray images. The
pre-operative DRRs and the intra-operative images are registered
and merged to provide both structural and functional information
about the tumor and the effected organ. Subsequent images taken
intra-operatively using the X-ray system can then be merged with
the pre-operative image over time to assist the physician. The
pre-operative images can provide detail about the anatomy that is
the subject of the procedure. Three dimensional image modalities
such as Computed Tomography (CT) and Magnetic Resonance Imaging
(MRI) contain high resolution information about the imaged
anatomy.
[0004] The intra-operative images are typically two dimensional
images that are available to provide the physician with an
indication of the current state of the anatomy in question. X-ray
and fluoroscopy images are typically used for these purposes. Two
dimensional (2D) images take significantly less time to acquire
than three dimensional (3D) images and are less intrusive to the
physician. However the resolution and detail of the 2D images are
inferior to that of its 3D counterpart. By combining the
pre-operative and intra-operative images by registering the two
images, the physician can have the benefit of the detail of the
pre-operative images and the current state of the patient via the
intra-operative images.
[0005] Conventional registration of a projection image to a
volumetric data set involves three steps. First, computation of a
simulated projection image (e.g., Digitally Reconstructed
Radiographs (DRRs)) is performed given the current relative
position of an X-ray source image and the volume. Second,
computation of the similarity measure and/or difference measure
quantifying a metric for comparing the X-ray or portal image to the
DRR is performed. Third, an optimization scheme is employed which
searches through the parameter space (e.g., six dimensional rigid
body motion) in order to maximize the similarity measure or
minimize the difference measure. Once the optimum position is
found, the DRR image should match the X-ray image.
[0006] The registration of two dimensional (2D) and three
dimensional (3D) images is a well-known technique. It is important
to compute the DRR so that it matches the real X-ray image in terms
of both brightness and contrast. In addition, a well-behaved
similarity measure should be chosen that can robustly characterize
a metric for the images. In order to make such an algorithm
practical, the computational time has to be reduced. Based on the
current state of the art, implementation of such techniques for
typical 3D volume data sets have a computation time of a few
minutes. Most of the computation time is spent on generating DRRs.
Another factor affecting the computation time is the number of
iterations that have to be computed.
[0007] One approach for reducing the computation time is to
randomly sample the DRRs and only use those samples for performing
computations, thereby reducing the computational complexity.
However, one drawback to this approach is that the robustness of
the results is compromised since less information is available to
the optimizer to take an accurate step toward the global solution.
For many practical applications, especially interventional
scenarios, registration time is crucial. It would be desirable to
be able to perform registrations in real-time or close to
real-time.
SUMMARY OF THE INVENTION
[0008] The present invention is directed to a system and method for
registering pre-operative images of an object with an
intra-operative image of the object. Prior to an operative
procedure, Digitally Reconstructed Radiographs (DRRs) are generated
for the pre-operative images of each individual patient. Signatures
are extracted from the DRRs. The signatures are stored in a
knowledge base. During the operative procedure, a signature is
extracted from the intra-operative image. The intra-operative
signature is compared to the stored pre-operative signatures. A
pre-operative image having a best signature match to the
intra-operative signature is retrieved. The retrieved pre-operative
image is registered with the intra-operative image.
BRIEF DESCRIPTION OF THE DRAWINGS
[0009] Preferred embodiments of the present invention will be
described below in more detail, wherein like reference numerals
indicate like elements, with reference to the accompanying
drawings:
[0010] FIG. 1 is a schematic block diagram of an illustrative
system for implementing a method of registering pre-operative
images with intra-operative images in accordance with the present
invention;
[0011] FIG. 2 is a flow diagram illustrating a method for
registering pre-operative images with intra-operative images in
accordance with the present invention; and
[0012] FIG. 3 illustrates a method for registering a pre-operative
and intra-operative image in accordance with the present
invention.
DETAILED DESCRIPTION
[0013] The present invention is directed to a method for
registering pre-operative high resolution images with
intra-operative low resolution images. In accordance with the
present invention, Digitally Reconstructed Radiographs (DRRs) at
various poses are precomputed for the pre-operative images of each
individual patient and stored with the images in a database.
Because the DRRs are computed prior to the interventional procedure
most of the complexity and computation time has been eliminated
from the typical registration process. As such, registration of a
pre-operative and intra-operative image can be processed in an
extremely fast and efficient manner. Prior art registration
techniques can take up to a minute to compute. By implementing the
method of the present invention, registration can be accomplished
in real-time (up to 20 frame/s).
[0014] FIG. 1 illustrates a schematic block diagram of an
illustrative system for implementing the method of the present
invention. The present invention exploits the pre-operative images
and intra-operative images to provide a more useful and inexpensive
registered image of an organ, which is the subject of a minimally
invasive therapeutic intervention. For example, a tumor can be
imaged both pre-operatively using a three dimensional imaging
system, such a Computed Tomography (CT) system or a Magnetic
Resonance Imaging (MRI) system and inter-operatively using a two
dimensional imaging system such as an X-ray or fluoroscopy device.
The images are registered and merged to provide both structural and
functional information about the tumor and the effected organ.
Subsequent images taken intra-operatively can then be merged with
the pre-operative image over time to assist the physician.
[0015] In accordance with the present invention, a database 110 is
generated based on the pre-operative volumetric data. The
constructed database 110 includes the position information (e.g.,
pose) of each entry and also a set of signatures or features
pertaining to the particular image, which are extracted from the
digitally reconstructed radiographs (i.e., DRR) at the same
position. Examples of such signatures include intensity histogram,
invariants derived from multiscale Gaussian filters or Gabor
filters, the image itself and so on. These features could be used
in a hierarchical manner ordered by their query efficiency. As will
be described in more detail hereinafter, matching of the
interventional (intra-operative) image space to the pre-operative
diagnostic space is then reduced to pose and retrieval of the
extracted signature of the current projection image within the
database 110.
[0016] An image of a desired tissue region or an organ is obtained
by employing an imaging system 102 such as, for example, a CT or
MRI device. Data is collected for images of the tissue region or
organ and stored for further processing by processor 108. These
images are obtained prior to any operative procedure. Other organs
or internal structures may also be imaged as needed. The images are
then reconstructed into DRRs and the above knowledge base is
generated and stored in database 110 for each DRR.
[0017] Images of the same desired tissue region or organ are then
obtained by employing an intra-operative imaging system 104 which
may also be an X-ray device or linear accelerator. During the
operative procedure, an initial image is obtained and stored in
processor 108. Rigid registration of the image from the
intra-operative imaging system and the images from the
pre-operative imaging system is performed. Preferably, the image
taken by the intra-operative imaging system is matched to an image
taken pre-operatively that has the same pose and is in a relatively
similar state. For example, an internal organ that is imaged should
be in approximately the same state for both imaging processes to
ensure proper registration. Identifying pre-operative images having
the same pose and signatures is greatly simplified because a match
can easily be found by performing a look up in the knowledge
base.
[0018] Two scenarios can be considered. First in abdominal and
thoracic procedures, the rigidity of the internal organ movement
can be assured using either breath-hold techniques or gating
techniques (e.g., both the pre-procedural and first set of
inter-procedural image is taken at the full inhalation). Second for
neurosurgical procedures, only after craniotomy, there exists some
deformable movement of the structure, which is so-called
brain-shift. Therefore, the rigidity assumption for this stage is
quite reasonable.
[0019] As indicated above, the image data from the
preoperative-imaging system 102 and the intra-operative imaging
system 104 are input to processor 108. Processor 108 may include a
Graphical User Interface (GUI), which permits a user to manually
draw a border or contour around a region of interest in the images.
Alternatively, a segmentation algorithm may be employed to
differentiate regions of interest and draw contours for images
without user interaction. Segmentation algorithms known to those
skilled in the art may be employed. Database 110 stores the
images.
[0020] A display 106 is included for displaying the images and
displaying the registered images. An interface device or devices
112 are also included such as a keyboard, mouse or other devices
known in the art.
[0021] FIG. 2 is a flow chart that illustrates an exemplary method
for registering pre-operative images with intra-operative images in
accordance with the present invention. Pre-operative images of a
patient are obtained using a three dimensional (3D) imaging system
(step 202). A knowledge base is generated from the pre-operative
volumetric data set associated with the 3D images (step 204). The
knowledge base comprises thousands of DRR images that are computed
from the volume for all the specified poses using standard DRR
algorithms implemented either by software or by hardware. Each pose
is defined by the projection matrix that mimics the actual
projection matrix of the targeted C-Arm system for 2D fluoro image
acquisition. The pose is varied with a specified sampling
resolution such that all the orientations in 3D are sampled
uniformly as much as possible.
[0022] The knowledge base has entries encoding information about
the DRR at various poses and the actual position and orientation of
the object being imaged. Typically these images are used in medical
applications for examining various human organs for medical
conditions or tumors. The type of information that is extracted
from the DRRs at each pose may vary depending upon the application.
It is also possible to save the whole DRR image along with the pose
information as one data entry. Each data entry in the knowledge
base is a compact representation of a DRR at a certain pose.
Features or signatures of the DRR can also be stored such as the
maximum intensity marginals of the image.
[0023] The knowledge base is arranged in a tree-like structure and
is arranged based on the position and orientation information in
which the neighborhoods can be defined. In accordance with the
present invention, the knowledge base is set up based on pose.
Similar poses are assigned to the same neighborhood. This will make
the retrieval of an entry at the corresponding pose faster and
easier.
[0024] The knowledge base has to be large enough to cover
discrepancies of the parameter space up to a certain degree. The
larger the coverage of the knowledge base in terms of pose, the
larger the operating base of the algorithm as a whole. The spacing
and/or resolution of the poses stored in the knowledge base can
vary depending upon the application. One implementation may
consider having a more compact knowledge base at the expense of
larger spacing among the poses (i.e., lower pose resolution). This
kind of implementation is then coupled with a refinement step where
a conventional approach can be used to achieve further accuracy and
to compensate for the lost resolution in the knowledge base.
[0025] Next intra-operative images are taken of the same patient
(step 206). An X-ray or fluoroscopy device is used to take the
images. Signatures of the image are extracted in the same way that
that knowledge base was generated (step 208). A metric is defined
to provide a distance measure among the data entries and the given
intra-operative image. The definition of the distance measure
depends upon the way that the signatures are defined. For simple
metric such as histogram, a sum of squared distance could be used,
whereas for image metric derived from Gaussian and Gabor filtering,
one can use distance measure derived from cross correlation and
mutual information of two images. Data retrieval is performed
through the knowledge base in order to retrieve the best signature
match of a pre-operative image with the current intra-operative
image (step 210). The pose of the resultant entry would carry the
registration information. Since the generation of the knowledge
base is computed prior to obtaining the intra-operative images,
extraction of the signatures of the X-ray images and search and
retrieval of the volumetric data in the database can be performed
very efficiently.
[0026] FIG. 3 illustrates the method of registering pre-operative
images with intra-operative images in accordance with the present
invention. Prior to a medical procedure, 3D pre-operative images
302 of the patient, in this case carotid vessels, are obtained in
the manner described above. DRRs are pre-calculated at various
poses for the images and stored in database 306. Next during the
medical procedure, 2D intra-operative images 304 of the carotid
vessels are taken. Preferably, the positioning and angle in which
the intra-operative images are taken are the same as those for the
pre-operative images. These images are also sent to database 306.
Next, the pre-operative images are retrieved from the database and
a real-time online registration of the pre-operative and
intra-operative images occurs. The resulting registered image 308
is displayed which is essentially a fused version of the 2D and 3D
images.
[0027] Having described embodiments for a method for registering
pre-operative DRRs with intra-operative images, it is noted that
modifications and variations can be made by persons skilled in the
art in light of the above teachings. It is therefore to be
understood that changes may be made in the particular embodiments
of the invention disclosed which are within the scope and spirit of
the invention as defined by the appended claims. Having thus
described the invention with the details and particularity required
by the patent laws, what is claimed and desired protected by
Letters Patent is set forth in the appended claims.
* * * * *